Well it depends on how you use that 70%. I split it in 10 trades, so each trade gets 1/10th of the total capital. Probability of total ruin is essentially zero since I'm long or short with no leverage but a batch of trades still may all be losing (that's how probability works). I can split it in more trades and rely on the central limit theorem to always win, but then the return goes to shit. If I put too little trades in a batch, I run into Gambler's fallacy, when "in which the roulette wheel spun black 26 times in succession at the Monte Carlo Casino". Even with perfect probabilities it's still within the realm of possible that you'll get unlucky many times. So I put "calculate" there since once you know the probability, it's no longer an experiment but a calculation which I was too lazy to make. 70% is "good". As long as that's real and actually happening, it's literally impossible to be so dumb to lose money on the long term.
First start with the statistics law which says that given a probability distribution with variables of mean mu = positive and variance sigma, summing such variables results in the mean growing linearly but the variance quadratically, in layman terms averaging enough samples you get definite positive mean exceeding the variance. With that, I can choose how many trades to average and in this case it's 10. So I allocate a tenth of capital to each trade in a batch of 10 before the next batch. The batches can also be serial in time, like in different days, or parallel, all in a single day. It's not the best number probably but seems to work well enough. Too small a batch size and variance ("drawdown" as you call it) becomes too high. Too high and the nice, ideal laws of statistics don't apply anymore, by various reasons like correlation between stocks but even if I remove that, the market is not a perfect statistics. I found out that theory works at "back of the napkin" level but once you try to get too rigorous, it breaks apart. Like in theory adding 100, 1000 trades in a batch should result in always highly positive mean. Well, it doesn't. It increases for a while then it reaches an asymptote, and that's ignoring transaction costs, which mean that eventually you start losing money.
if it works stick with it obvious. I'll stay with what most, if not all, others use - actual peak to trough.